Music Recommender systems and popularity

Music recommender systems are an integral part of modern-day music listening. They are adopted by streaming services such as Spotify and Apple Music and are part of their core feature set. These recommender systems, however, are not perfect. One of the issues described in “Support the underground: characteristics of beyond-mainstream music listeners” by Kowald et al. (2021) is the popularity bias these recommender systems have: popular songs are over-represented in recommendation lists. Furthermore, a study by Tintarev et al. (2013) found that openness to new music and the diversity of music within a group influences recommendation quality. The authors set to find out what the characteristics of beyond-mainstream music tracks and music listeners are, and how these characteristics correlate with openness and diversity patterns as well as with recommendation quality.

The authors of the article aimed to identify the characteristics of beyond-mainstream music tracks and listeners and how these characteristics influence recommendation quality in addition to openness and diversity patterns. The authors observed that beyond-mainstream music listeners receive worse recommendations, except for a certain subgroup with a high openness value. This subgroup received even better recommendations than mainstream music listeners.

Market concentration

Over the last couple of decades, music consumption has shifted from purchasing individual records and listening to the radio to a single digital application that fits into your pocket. This era of digitalization is dominated by a fistful of firms and platforms, that manage to concentrate the market and therefore obtain massive potential leverage. Luis Aguiar and Joel Waldfogel (2021) aim to study the effect of such influence by digital companies by using the case of Spotify in their article Platforms, Power, and Promotion: Evidence from Spotify Playlists. The authors use Spotify as an example as it, combined with Apple Music, represents almost 55% of all global music streaming subscriptions. With such market concentration, the potential for antitrust problems arises. Therefore, the authors wish to quantify Spotify’s market power and evaluate if further scrutiny is needed when dealing with these platforms. They particularly focus on Spotify’s playlists, as the company manages 84.3% of the top 1000 playlists on the platform. Given that Spotify shares revenue with creators, list inclusion could present an avenue for favoritism. The fact that Spotify also cuts licensing deals with independent artists and directly competes with labels that rely on the platform presents a conflict of interest and exacerbates the potential for shady business practices. Therefore, an analysis of Spotify’s market power is conducted by comparing periods of song inclusion and removal from their playlists and the changes in daily listening numbers that ensue. They do this using both curated and algorithmic playlists and discontinuous daily listening numbers that Spotify publishes. Curated playlists are handpicked by experts at Spotify, while algorithmic playlists follow a simple rule for inclusion.

First, they studied curated playlists like Top Hits Daily and RapCaviar. Using regression discontinuity, they figured out that a curated inclusion on the Today’s Top Hits playlist on average raises worldwide daily streams by approximately 260,000, which according to their estimates corresponds to an approximate total increase of 20 million with a $77,000 payout. A comparable effect in the opposite direction was also noticed when a song was removed from the playlist, as daily streams dropped.

A similar effect appeared with algorithmic playlists. The Global Top 50 playlist simply includes the 50 most streamed songs during the previous day. Using data for the top 200 most streamed songs, the authors constructed a fictitious 51st song and compared the cutoff. The songs decrease in daily streams the further you go down the playlist. They spotted a significant increase in the percentage reduction of daily streams per song ranking at rank 51. This change then disappears from 51 onwards, which signifies an important effect of inclusion on playlists, even if it is simply algorithmic.

The authors also studied a special playlist called New Music Friday, as songs on there are released just days before their inclusion. Therefore, consistent pre-inclusion listening data is not available. They figured out that playlist rank is important, as songs in the first spot have an 85% chance of appearing on chart playlists. That percentage drops drastically until it is effectively irrelevant around rank 11-15. Using data from this playlist, they also discovered that effects from song inclusion in playlists persist for circa 1000 days.

In the end, song inclusion in playlists has immense ramifications on streaming numbers and payouts. Spotify’s dominant position in the playlist selection therefore presents a problem for other market participants, which needs to be more heavily analyzed by antitrust authorities